The OpenMP API specification for parallel programming
Webinar: Using OpenMP to Harness GPUs for Core-Collapse Supernova Simulations with GenASiS
This webinar is about how to use OpenMP to harness the awesome computational power of GPUs for scientific simulations of core-collapse supernovas with GenASiS. Presented by Dr. Reuben D. Budiardja of the National Center for Computational Sciences at Oak Ridge National Laboratory.
In this talk, he describes efforts in harnessing the GPUs in leadership-computing platforms for simulations of core-collapse supernovas with the ORNL code, GenASiS (General Astrophysics Simulation System). Using OpenMP they have implemented a generic, portable infrastructure in GenASiS that greatly facilitates efficient interactions with the GPUs, and potentially with other hardware accelerators as supported by OpenMP compilers.
The presentation will start with the motivation for this work and a high-level overview of GenASiS, a program they are developing for large-scale simulations of astrophysical phenomena. Then he will describe how they use OpenMP runtime library routines with their storage classes to seamlessly and explicitly manage data movement and data association between the host and the device copy. He will also show how they use OpenMP directives to speed up computations, then share the results of their work on leadership computers at ORNL. The webinar will conclude with their current efforts to incorporate OpenMP metadirectives into their kernels for even better portability.
Who should attend?
Application developers wanting to learn new techniques in using OpenMP offload.
Compiler and standard developers who would like to see how an application uses OpenMP and see some example kernels that may benefit further optimizations.